As man portage and emerge --help now refer to actions with --action, I changed all the actions to _arguments. Now, emerge --ask -d --unmerge will complete the right set of packages, whereas before, the action had to come first. Added all the new arguments from new portage versions (--newuse, --fetch-all-uri, --metadata, --version, --getbinpkg, --getbinpkgonly). I'm not sure what to do about --getbinpkg{,only}, so it just completes the default package list. This patch removes the cache support. Instead, completion searches for directories and ebuilds matching $PREFIX wherever possible. After "emerge \=<tab>", only categories will get completed. If one letter is typed, all categories and ebuilds will be completed that start with said letter. On my computer(1800+ athlon), cache is a tiny bit faster at "emerge \=g<tab>", but as soon as more characters are typed, $PREFIX matching becomes just as fast. I removed cache because I feel it does not provide enough benifit over live searching. After an "emerge --sync", parts of the cache are wrong and the whole thing needs to be rebuilt for accuracy. Instead of sourcing directly, or portageq, I copied what /etc/profile does for grabbing variables. All variables are kept local and it's as fast as sourcing. Also, previously, environment variables weren't handled correctly. Setting PORTDIR in the environment would give the correct completions for the first <tab>, but the second <tab> would default to what PORTDIR is set to in make.{conf,globals}. Removed _portage_clean and _portage_quickpkg as _portage_unmerge{,_vers} are able to handle both tasks. Well actually, _portage_clean had exact versions without \= syntax. Now to clean a specific app, a \= is required(portage understands both \=pkg-ver and pkg-ver for --clean). IMHO, it's confusing to have different syntax for --unmerge and --clean, I chose the more standard \= syntax. _portage_clean could certainly be added back if it's functionality is wanted. Also, "emerge --unmerge <tab>" doesn't complete any versions. The standard =, >, etc.. are required now. Again, I changed this for consistency with the rest of portage(and so quickpkg doesn't require it's own function). Pretty much all actions that can accept arguments now do. --depclean, --sync, --search and --help now all have the appropriate arguments. General fixes: (kv) refers to an associative array, and we have none. _tags -s: -s is not an argument to _tags. Removed extra >,=,<=, etc.. logic. Changed scalars into arrays where appropriate. Sets and packages will no longer get completed on the same line, so no "emerge world <tab>" and get packages. --inject now works properly, although it requires some extra code. I'll be happy to see it gone. "emerge app-shell/zsh-completion/<tab>" completes "files". As far as I can tell, _path_files is incapabale of filtering out just "files"(and not miscfiles,etc..), so I used an ad-hoc compadd setup. _multi_parts is too slow for so many packages, and _sep_parts erroneously removes part of $PREFIX occasionally. Now that _path_files is gone, overlays have cat/pkg support. Added cat/pkg completion for \=,>,etc.. context. _sep_parts works fine here, AFAICT. Removed rsync. Moved all emerge related stuff into emerge(){}. So calling ebuild<tab> doesn't need to figure out all the packages, or emerge arguments. Removed whitespace from the end of a few lines. Bugs with new version: emerge -C<tab> completes all the options. -C was handled by _values previously, and now _arguments, so I'm not sure if this can or should be fixed. emerge --oneshot completes sets(world system). AFAICT, _arguments is incapable of excluding options provided by _values, so the only fix I see would be making a special case for --oneshot similar to how --unmerge, world and --usepkgonly are handled. Personally, I think it's a minor bug and not worth the hassle as portage doesn't complain. Those are the only two bugs I've found with the current patch, but it changes alot of code, so I might have broken something. "rc-update del" only completes init-scripts that are in a runlevel. Changed "add" to _values to be consistent in appearance with "del". Also, ignore files that end in .sh. Well, I think that's it. Reproducible: Always Steps to Reproduce: 1. 2. 3.
Created attachment 44159 [details, diff] patch to _portage
Thanks for the update. Yes, I should have added your name to authors list ;-) At first sight, it seems much cleaner than current one. As for cache support, I suspected that the cache actually accelerates completion speed, and thinks it's good to remove if it is not necessary. I generally agree with you on other things you mentioned. (I'll look in detail later) I'll look into it asap, but I'm afraid it will take at least a week as I'm moving out of my home right now.
The patch seems identical to _portage patch posted to bug #71957, and as I applied _portage patch attached there I close this bug. (tested several weeks so far without a problem) Thanks oberyno!
